Gravicode.MediaPipeNet.Models.Quantized
1.0.0
dotnet add package Gravicode.MediaPipeNet.Models.Quantized --version 1.0.0
NuGet\Install-Package Gravicode.MediaPipeNet.Models.Quantized -Version 1.0.0
<PackageReference Include="Gravicode.MediaPipeNet.Models.Quantized" Version="1.0.0" />
<PackageVersion Include="Gravicode.MediaPipeNet.Models.Quantized" Version="1.0.0" />
<PackageReference Include="Gravicode.MediaPipeNet.Models.Quantized" />
paket add Gravicode.MediaPipeNet.Models.Quantized --version 1.0.0
#r "nuget: Gravicode.MediaPipeNet.Models.Quantized, 1.0.0"
#:package Gravicode.MediaPipeNet.Models.Quantized@1.0.0
#addin nuget:?package=Gravicode.MediaPipeNet.Models.Quantized&version=1.0.0
#tool nuget:?package=Gravicode.MediaPipeNet.Models.Quantized&version=1.0.0
Gravicode.MediaPipeNet.Models.Quantized
Reduced-precision variants of the MediaPipe.NET models, copied to models/ in your output folder:
- FP16 — float16 weights and activations (inputs/outputs stay float32): half the size, fastest on GPUs (DirectML, CUDA). 14 models.
- INT8 — weight-only int8 (per-channel), computed in float: about a quarter of the size with near-identical results. 9 models.
Every variant was validated against its float32 model (cosine similarity ≥ 0.999 for FP16, ≥ 0.99 for INT8).
Select them with InferenceOptions.Precision; models without a variant use float32:
var options = new BaseOptions { Inference = new InferenceOptions { Precision = ModelPrecision.Float16 } };
using var landmarker = FaceLandmarker.Create(new() { BaseOptions = options });
MediaPipe.NET — a native .NET 10 port of Google MediaPipe's tasks on ONNX Runtime.
Created by Gravicode Studios, led by Kang Fadhil. Library: Apache-2.0. Model weights: © Google LLC, Apache-2.0.
Documentation (English & Bahasa Indonesia): see the docs/ folder of the repository.
Learn more about Target Frameworks and .NET Standard.
-
net10.0
- No dependencies.
NuGet packages
This package is not used by any NuGet packages.
GitHub repositories
This package is not used by any popular GitHub repositories.